Evaluation, Reproducibility, Benchmarks Meeting 13

Minutes of meeting 13

Date: 27th July 2021

Present: Carole, Annika, Lena, Nicola

---

TOP 1: MONAI open datasets

• See discussion from last meeting (June 23rd 2021)
• NCI offered to assist on hosting the data, but they are limited to specific applications which may it make difficult to host all available datasets
• Upcoming call with Prerna to discuss the details
  • Goals:
  • 1. Retrospective challenge organizer involvement
  • 2. Prospective support for future challenge organizers
    • It would be great if we could be ready for MICCAI 2022 challenges
  • Carole will be our representative (and link to the data WG)
  • Survey to incorporate community feedback on which platforms are currently used by challenge organizers to host the data (2-3 questions)
    • Invite MICCAI 2021 challenge organizers
    • Possible questions:
      • Are they interested in providing their datasets to MONAI?
      • Which platforms are they currently using and which ones would be okay to use (if they are interested in MONAI linking)?
    • Finalize and send beginning of september
  • Boundary conditions from MONAI:
    • It would be nice to find solutions that work for all challenges instead of having many very different APIs
    • Who is building the integration part? Probably MONAI should do as much as possible
  • A question may be added in the structured challenge submission system for MICCAI 2022 challenge submissions, asking whether they would be interested
    • generally to be link with MONAI
    • present one working solution and ask them whether this would be feasible for them
  • Open questions for Prerna:
    • Ask Prerna whether she could help regarding regulatory support
    • Which data hosting options are (not) possible from MONAI’s side?
    • How should we deal with more complex use cases? Example: Challenge data can only be used for academic purposes, not for companies and startups. How can we make sure this holds?
• Questions that should be added to the structured challenge submission system
  • Is registration needed to access data?
  • Data hosting platform

---

TOP 2: Taskforce report (metric implementation)

• Metric taskforce collected feedback regarding the metric implementation and tutorials and sent it to the developer team

---

TOP 3: Delphi process - metrics matter

• Annika won the best short oral presentation audience award at MIDL 2021
• Several companies approached Nicola and told that they really like the dynamic arXiv paper
• Intermediate questionnaires for expert groups
  • Goals:
    • Finalize the list of task-specific pitfalls: Based on the “problem characteristics” compiled by the consortium, generate a comprehensive list of potential pitfalls, including illustrations and references (extension of our ArXiv paper https://arxiv.org/abs/2104.05642)
    • Come up with a candidate list of medical/biological scenarios for which we want to provide concrete best practice recommendations in the paper.
  • Grouped problem properties into several properties
    • Reformulations:
      • Size of structures (e.g. size relative to pixel or image size)
      • Variability of sizes of structures (intra and inter image)
      • Acquisition protocols (e.g. angle or viewpoint)
      • Presence of artifacts or artificial structures, (such as text overlay) OR presence of image-modifying effects (such as artifacts or artificial structures)
    • Add comment box at the end of every category